vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

more cleanups

Browse source
This commit is contained in:
Olivier Chafik 2025-01-22 03:20:47 +00:00
parent 28cac497a6
commit 2dd09c792f
2 changed files with 8 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

26
common/sampling.cpp
View file

@ -151,22 +151,16 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
lparams.no_perf = params.no_perf; lparams.no_perf = params.no_perf;
std::vector<const char *> c_trigger_words; std::vector<const char *> trigger_words;
c_trigger_words.reserve(params.grammar_trigger_words.size()); trigger_words.reserve(params.grammar_trigger_words.size());
for (const auto & str : params.grammar_trigger_words) { for (const auto & str : params.grammar_trigger_words) {
c_trigger_words.push_back(str.c_str()); trigger_words.push_back(str.c_str());
} }
auto * result = new common_sampler { auto * result = new common_sampler {
/* .params = */ params, /* .params = */ params,
/* .grmr = */ llama_sampler_init_grammar( /* .grmr = */ llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root",
vocab, trigger_words.data(), trigger_words.size(),
params.grammar.c_str(), params.grammar_trigger_tokens.data(), params.grammar_trigger_tokens.size()),
"root",
c_trigger_words.data(),
c_trigger_words.size(),
params.grammar_trigger_tokens.data(),
params.grammar_trigger_tokens.size()
),
/* .chain = */ llama_sampler_chain_init(lparams), /* .chain = */ llama_sampler_chain_init(lparams),
/* .prev = */ ring_buffer<llama_token>(std::max(32, params.n_prev)), /* .prev = */ ring_buffer<llama_token>(std::max(32, params.n_prev)),
/* .cur = */ {}, /* .cur = */ {},
@ -237,9 +231,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
void common_sampler_free(struct common_sampler * gsmpl) { void common_sampler_free(struct common_sampler * gsmpl) {
if (gsmpl) { if (gsmpl) {
if (gsmpl->grmr) { llama_sampler_free(gsmpl->grmr);
llama_sampler_free(gsmpl->grmr);
}
llama_sampler_free(gsmpl->chain); llama_sampler_free(gsmpl->chain);
@ -258,9 +250,7 @@ void common_sampler_accept(struct common_sampler * gsmpl, llama_token token, boo
} }
void common_sampler_reset(struct common_sampler * gsmpl) { void common_sampler_reset(struct common_sampler * gsmpl) {
if (gsmpl->grmr) { llama_sampler_reset(gsmpl->grmr);
llama_sampler_reset(gsmpl->grmr);
}
llama_sampler_reset(gsmpl->chain); llama_sampler_reset(gsmpl->chain);
} }

1
src/llama-sampling.cpp
View file

@ -1504,7 +1504,6 @@ struct llama_sampler * llama_sampler_init_grammar(
size_t num_trigger_words, size_t num_trigger_words,
const llama_token * trigger_tokens, const llama_token * trigger_tokens,
size_t num_trigger_tokens) { size_t num_trigger_tokens) {
// struct llama_sampler * llama_sampler_init_grammar(const struct llama_vocab * vocab, const char * grammar_str, const char * grammar_root) {
auto * ctx = new llama_sampler_grammar; auto * ctx = new llama_sampler_grammar;
if (grammar_str != nullptr && grammar_str[0] != '\0') { if (grammar_str != nullptr && grammar_str[0] != '\0') {